tensorflow/models · error · NotImplementedError

Keras model.fit on TPUs is not implemented.

Error message

Keras model.fit on TPUs is not implemented.

What it means

Error "Keras model.fit on TPUs is not implemented." thrown in tensorflow/models.

Source

Thrown at official/legacy/transformer/transformer_main.py:335

          if params["enable_tensorboard"]:
            for metric_obj in train_metrics:
              tf.summary.scalar(metric_obj.name, metric_obj.result(),
                                current_step)
              summary_writer.flush()

        for cb in callbacks:
          cb.on_train_end()

        if flags_obj.enable_checkpointing:
          # avoid check-pointing when running for benchmarking.
          checkpoint_name = checkpoint.save(
              os.path.join(flags_obj.model_dir,
                           "ctl_step_{}.ckpt".format(current_step)))
          logging.info("Saved checkpoint to %s", checkpoint_name)
      else:
        if self.use_tpu:
          raise NotImplementedError(
              "Keras model.fit on TPUs is not implemented.")
        history = model.fit(
            train_ds,
            initial_epoch=current_iteration,
            epochs=current_iteration + 1,
            steps_per_epoch=train_steps_per_eval,
            callbacks=callbacks,
            # If TimeHistory is enabled, progress bar would be messy. Increase
            # the verbose level to get rid of it.
            verbose=(2 if flags_obj.enable_time_history else 1))
        current_step += train_steps_per_eval
        logging.info("Train history: {}".format(history.history))

      logging.info("End train iteration at global step:{}".format(current_step))

      if (flags_obj.bleu_source and flags_obj.bleu_ref):
        uncased_score, cased_score = self.eval()
        cased_score_history.append([current_iteration + 1, cased_score])

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/legacy/transformer/transformer_main.py:335 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/5053f6f6ecb669d0. Report an issue: GitHub.